Cluster Autoscaler

The Cluster Autoscaler adjusts the number of nodes in your cluster. It adds nodes when pods can't be scheduled (pending) and removes underutilized nodes whose pods can be rescheduled elsewhere.

How It Works

The Cluster Autoscaler scans every 10 seconds. For scale-up: it detects pods in Pending state, evaluates which node group can satisfy requirements, and calls the cloud provider API to provision new VMs. For scale-down: it checks if nodes are below 50% utilization for 10+ minutes and safely drains them.

It uses an expander strategy to choose which node group to scale: least-waste (recommended), most-pods, random, or priority.

When to Use

• Pod autoscalers (HPA/KEDA) create more pods than nodes can handle
• You want automatic infrastructure elasticity
• Significant load variance (daily cycles, seasonal spikes)
• Cost optimization by removing idle nodes

When NOT to Use

• Bare metal clusters (no dynamic node provisioning)
• Latency-sensitive workloads (2-5 min node provisioning)
• Fixed-capacity requirements or compliance constraints

Real-World Example

Step-by-Step Implementation (AWS EKS)

1. Configure node group

aws eks create-nodegroup \
  --cluster-name my-cluster \
  --nodegroup-name standard-workers \
  --instance-types c5.2xlarge \
  --scaling-config minSize=2,maxSize=100,desiredSize=5

2. Deploy via Helm

helm repo add autoscaler https://kubernetes.github.io/autoscaler
helm install cluster-autoscaler autoscaler/cluster-autoscaler \
  --namespace kube-system \
  --set autoDiscovery.clusterName=my-cluster \
  --set awsRegion=us-east-1 \
  --set extraArgs.expander=least-waste \
  --set extraArgs.scale-down-unneeded-time=10m \
  --set extraArgs.scale-down-utilization-threshold=0.5

3. Verify

kubectl logs -n kube-system -l app=cluster-autoscaler --tail=50
kubectl get nodes --watch

Common Pitfalls